Academia.eduAcademia.edu

Figure 9 – uploaded by Karim Bouamrane

See full PDFdownloadDownload figure
TABLE 5. Results of the hyper-heuristic and the VNS algorithms for medium and large size instances.

Table 5 Results of the hyper-heuristic and the VNS algorithms for medium and large size instances.

Related Figures (20)

FIGuRE 1. A typical hub-and-spoke based round-trip service network.
FIGuRE 1. A typical hub-and-spoke based round-trip service network.
FIGURE 2. Network structure for a hub-and-spoke structure distribution network of humani- tarian aids in Lebanon. Source: Displaced Syrian distribution by CAZA, Lebanese Government.
FIGURE 2. Network structure for a hub-and-spoke structure distribution network of humani- tarian aids in Lebanon. Source: Displaced Syrian distribution by CAZA, Lebanese Government.
FIGURE 3. A typical structure of network in a runaway hub-and-spoke structure.
FIGURE 3. A typical structure of network in a runaway hub-and-spoke structure.
TABLE 1. A summary of the relevant contributions in the literature. the hub nodes. In contrast to a backup hub node and sophisticated hub edges (connecting such backup node to the hub-level network) which are expensive to setup and maintain, the runway nodes are some enhancements to the normal spoke nodes and are only connected to one or two peer runaway nodes using connections that are somewhere between and maintain until the d occurrence and becomes the performance of a hub edge and a spoke edge, making it much cheaper to operate isruption is over. Moreover, when the hub-level network is targeted by a disruptive temporarily unavailable, this notion provides excess capacity somewhere outside the focus of disruption and in a decentralized fashion — i.e. at the spoke level network. We have carried out a thorough investigation o relevant decision makers f the matter in two cases from the real practice and a close collaboration with the rom the sectors. From the theoretical and modeling point of view, this model opens a new perspective in includ ing redundancy in the hub-and-spoke structures and generalizes some of the previous work (including our work in [21]). We propose the first (exponential) mixed integer programming formulation for this problem, an efficient separation routine to separate from among the exponentially many constraints, and an efficient heuristic-selection hyper-heuristic delivering high quality solutions in a very reasonable time. Extensive computational experiments on randomly generated instances of various sizes, confirm computational efficiency of the proposed solution framework and the viability of the approach.
TABLE 1. A summary of the relevant contributions in the literature. the hub nodes. In contrast to a backup hub node and sophisticated hub edges (connecting such backup node to the hub-level network) which are expensive to setup and maintain, the runway nodes are some enhancements to the normal spoke nodes and are only connected to one or two peer runaway nodes using connections that are somewhere between and maintain until the d occurrence and becomes the performance of a hub edge and a spoke edge, making it much cheaper to operate isruption is over. Moreover, when the hub-level network is targeted by a disruptive temporarily unavailable, this notion provides excess capacity somewhere outside the focus of disruption and in a decentralized fashion — i.e. at the spoke level network. We have carried out a thorough investigation o relevant decision makers f the matter in two cases from the real practice and a close collaboration with the rom the sectors. From the theoretical and modeling point of view, this model opens a new perspective in includ ing redundancy in the hub-and-spoke structures and generalizes some of the previous work (including our work in [21]). We propose the first (exponential) mixed integer programming formulation for this problem, an efficient separation routine to separate from among the exponentially many constraints, and an efficient heuristic-selection hyper-heuristic delivering high quality solutions in a very reasonable time. Extensive computational experiments on randomly generated instances of various sizes, confirm computational efficiency of the proposed solution framework and the viability of the approach.
FIGURE 5. pHLPwR without the network structure (p = 4 clusters).
FIGURE 5. pHLPwR without the network structure (p = 4 clusters).
Once Algorithm 1 terminates, we inter-connect all the hubs (complete graph) and construct the cycles for each cluster using a greedy constructive heuristic choosing the cheapest edge to add at every step. Our extensive preliminary computational experiments has revealed that this method is effective and provides quality initial solution for our hyper-heuristic.
Once Algorithm 1 terminates, we inter-connect all the hubs (complete graph) and construct the cycles for each cluster using a greedy constructive heuristic choosing the cheapest edge to add at every step. Our extensive preliminary computational experiments has revealed that this method is effective and provides quality initial solution for our hyper-heuristic.
TABLE 2. The parameters used in Algorithm 3.
TABLE 2. The parameters used in Algorithm 3.
TABLE 3. Computational results of MIP-HLPwR. the “GAP” column is calculated using the best-known solution of the hyper-heuristic and best-known solution of VNS (a negative GAP means that VNS gave a better solution than the hyper-heuristic).
TABLE 3. Computational results of MIP-HLPwR. the “GAP” column is calculated using the best-known solution of the hyper-heuristic and best-known solution of VNS (a negative GAP means that VNS gave a better solution than the hyper-heuristic).
FIGURE 7. Numerical results of the MIP. (a) MIP time. (b) Objetive value comparison.
FIGURE 7. Numerical results of the MIP. (a) MIP time. (b) Objetive value comparison.
TABLE 4. Results of the hyper-heuristic and the VNS algorithms for small size instances.
TABLE 4. Results of the hyper-heuristic and the VNS algorithms for small size instances.
TABLE 5. continued. TABLE 6. Hyper-heuristic structures information for medium and large size instances.
TABLE 5. continued. TABLE 6. Hyper-heuristic structures information for medium and large size instances.
FIGURE 8. Numerical results of the H-HLPwR. (a) Execution time. (b) Objective value.
FIGURE 8. Numerical results of the H-HLPwR. (a) Execution time. (b) Objective value.
Related topics:
Information SystemsApplied Mathematics
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved